Efficient Post-Shrinkage Estimation Strategies in High-Dimensional Cox's Proportional Hazards Models

Area of Science:

• Biostatistics
• Statistical Modeling
• Survival Analysis

Background:

• Standard regularization methods (LASSO, Elastic-Net, SCAD) excel at variable selection but often neglect weaker signals, potentially biasing parameter estimates.
• Existing corrected shrinkage estimators are limited to linear models, leaving a gap in their application to survival data, which frequently involves both strong and weak effects.

Purpose of the Study:

• To develop and evaluate a novel class of post-selection shrinkage estimators specifically designed for the Cox proportional hazards model.
• To address the limitation of existing methods in handling weak signals within survival regression frameworks.
• To improve estimation and prediction accuracy in survival analysis by incorporating both strong and weak effect variables.

Main Methods:

• Proposed a new family of post-selection shrinkage estimators tailored for the Cox model.
• Established the asymptotic properties of the newly developed estimators.
• Conducted simulation studies incorporating weak signals to assess performance.
• Validated the approach on two real-world biomedical datasets.

Main Results:

• The proposed shrinkage estimators demonstrated the potential to enhance estimation accuracy by effectively incorporating weak signals.
• Simulations confirmed improved prediction accuracy compared to existing methods when weak signals were present.
• Application to real-world datasets validated the practical utility and advantages of the novel approach.

Conclusions:

• The novel post-selection shrinkage estimators offer a significant advancement for variable selection in Cox regression models.
• This method effectively addresses the challenge of weak signals, leading to more robust and accurate statistical modeling in survival analysis.
• The findings have broad implications for biomedical research where accurate identification of risk factors is crucial.